How far are we from autonomous drone swarms, where a single operator can designate a killzone, like on an rts game and have drones look up targets, engage them and then come back home to recharge and reload?
How far are we from autonomous drone swarms, where a single operator can designate a killzone, like on an rts game and have drones look up targets, en...
weve been there for years
a hobbyist can build one at home with off the shelf aliexpress components. a arducopter supporting flight controller and a jetson nano running a classification model to detect people from the camera feed.
its not even expensive
exactly what I've thought - that something like this should be possible. yet, when you look at the ukrainian war, they are still manually controlling the drones and talking about the importance of training new drone operators.
If a single person could operate hundreds or thousands of drones, it would seem like a cheap way to deal with an opponent with higher number of troops.
Off the top of my head, i remember an average US soldier costing USD 7m to train and equip - that would get you a whole bunch of cheap chinese drones and grenades
>yet, when you look at the ukrainian war, they are still manually controlling the drones
Part of the problem has been the Ukrainian military command. The guys on top were trained in the old Soviet model of "we artillery other side until it surrender, great success!"
That one top brass who got fired a couple of months ago, I remember reading commentary from multiple Ukrainian sources immediately afterward saying they were glad the stupid fuck was thrown out because he had been blocking effective drone programs for the previous 18 months. Wasn't really clear whether it was because he was on Putin's payroll or just refused to believe that drones could be anything more than toys.
First off, Ukraine and Russia are 1980 militaries with a few modern toys
Second off, Drones can be jammed and "cheap" drone "swarms" are more vulnerable to every conceivable countermeasure because they are packed more densely.
Sending seven million dollars worth of Amazon drones against every soldier is just going to waste time and money you could have spent not getting hit with a real big boy airstrike.
No we’re not there at all. AI is nowhere close to being good enough for autonomous drones. The best we have is backup systems that target a pre designated area but they can essentially be visually spoofed and electronically jammed still. Which is why those backup systems are hard coded not “thinking” AI. EW is the biggest thing that will hold drones back and AI probably won’t matter for a very long time if ever. It takes a lot of energy just to run a chatbot and it needs hundreds of millions of databases to pull from to not tell you orange banana when you ask it what 3+3 is, databases that won’t be possible to rely on in a heavily jammed airspace. And won’t be possible to fit on a disposable kamikaze drone.
You definitely do NOT need an active network connection to allow a computer to distinguish human objects with relative ease, you do not need terabytes and terabytes of disk space or hundreds of watts of electricity for CPU/GPU (that’s more for training the model, once the model is trained it’s pretty computationally cheap to use). Your smartphone can run these calculations entirely on its own hardware.
Sure they can be spoofed with dummy humans I guess but it doesn’t need to be 100% accurate. They can also use thermal cameras in conjunction with optical recognition to aid in target designation.
Arm a drone with a small caliber gun (since I would imagine recoil would be a concern for anything flying), develop a way to lock on to ground targets and fire accurate shots while moving, and you have a pretty terrifying system. If you can mass produce them by the thousands and give them decent range and speed it’ll be a fucking nightmare even if they’re firing the equivalent of 22lr projectiles.
You can spoof an AI that detects humans by doing jumping jacks or literally just wearing a cape.
GPS doesn’t work if you’re being jammed. It will work but hit the wrong area if the GPS is being spoofed.
AI can spot targets based on thermal signature but won’t be able to differentiate a military thermal and a civilian thermal or a drum fire reliably no. Again, AI doesn’t “think” we don’t have real AI right now. The countermeasures to AI are extremely cheap and thus make any system reliant on them worse than just human operated. To say nothing of the problems of politics if an autonomous drone wipes out a wedding or kiss at a bonfire. We have had drones for a while capable of autonomous kill, but they only get used in places like the DMZ where they still have to have a human press the trigger. Because AI has trouble distinguishing a human from wildlife sure but also because if it just shoots at everything without an IFF it may blue on blue someone with a wrong/faulty IFF or a defector far more often than it’ll kill a Nork infiltrator. Friendly fire from AI is extremely dangerous and extremely likely with how limited and just, bad, current AI detection algorithms are. They’re decades away from even matching as poor a track record as A10 pilots, throwing hunter killers out now would be like giving WW2 typhoon pilots cocaine and telling them to go wild.
Good AI won’t get spoofed by mere jumping jacks and especially not if it uses heat signatures to aid it.
You’re right about the issue of collateral damage or friendly fire which is why we’re probably not using them yet. I would envision this as some kind of area denial weapon akin to a minefield except you can drop it instantaneously wherever you want. You could just designate a rectangular kill zone on a map where the drones fly to and shoot anything that resembles a human. It doesn’t need to be 100% effective at identifying targets, it would be something to harass the enemy with and force them to seek cover or do weird shit like jumping jacks to fool the AI that would reduce their combat effectiveness.
>Good AI
AI that actually exists, on the other hand...
The current technology is more than up to the task of identifying humans. It’s not “intelligent”, AI is just a buzzword, but it works
It is, but it can be spoofed by all kinds of things which isn't a problem unless you're sending a huge drone swarm and they all get spoofed into one cheap decoy
You could even have the drones go into some kind of reduced power sentry mode where they land and just scan the area for targets, maybe give them solar panels that provide enough juice to run the computer and sensors, then when enemies are detected they spin up and go into attack mode. None of this sounds impossible with the current technology that we have but it would make warfare a fucking nightmare so I hope we don’t see it any time soon
The best AI in the world has failed 100% when put under stress testing by real humans trying to get around it. The problem is that AI is not AI, it’s recognition models. If something not in it’s database shows up it cannot identify it because it doesn’t think, it’s not AI, it’s just matching pictures. The reason hunter killers aren’t used is because a literal cardboard box can defeat them.
I wonder how effective a system with a combination of GPS, visual image analysis, and motion sensors would be at determining where a drone is. I am pretty sure stuff like that exists but I'm unsure of how small of a package you could the required hardware into
>AI can't navigate aircraft to a location based on GPS data
>AI can't spot targets based on thermal signatures and drop a payload when directly overhead
I think the main concern is practicality, not possibility. If you still need a drone operator to deploy it, there's not really a ton of reason to overcomplicate things. Its better to give 20 guys one drone they can directly control than give one guy an autonomous drone swarm. Identifying human beings based on visual and thermal cameras isn't that difficult, but identifying what side they're on is more so. Obviously there are a lot of ethical concerns, and no country wants to be seen as the 'bad guy', its better to say 'oops human error XD' when a drone blows up a civilian than it is to say 'our autonomous system killed someone it should not have'
also this
And yet no military or insurgency has done so, despite ample opportunity, funding and technical talent being available. Ergo, it's not as easy as you imply.
Idfk why people think those toy drones are the same thing as military UAVs.
It’s like saying a shitbox is the same thing as a tank.
Sure you can make a car bomb out of the former but it being a legitimate weapon is dubious.
>weve been there for years
this is correct. MQ-9 was fully autonomous in 2007. The only reason the US only flies it remote on kill missions is doctrine.
Not far if that's really something that's worth developing. Assuming you mean rotary drones specifically, right now their problem is flight time and payload. Theoretically you can allow yourself a bigger size and weight criteria to compensate, but that also makes the drone a bigger target with even less evasive capability. Also, bigger rotors tend to make aircraft louder so the enemy will definitely hear it long before they see it. Another issue is weapon systems. A gun system tends to be pretty heavy and recoil is worse for rotary drones. With modern technology, it's possible to mitigate these, but you'll just end up with an expensive flying peashooter with questionable efficacy. For this reason, I personally feel loitering munition/kamikaze drones rigged with small explosives to be a better near-term solution for rotary attack drones over ones with mounted gun systems.
As far as AI decision-making goes, ethics is more immediate looming concern which is being debated by the enemies of progress as we speak. You will live to see man-made horrors beyond your comprehension. It's truly quite exciting.
>progress
Obviously there is much involved into this. A big worry is collateral damage. In addition, automizing them may not work super well. You could have an army of drone operators do it. Which would be a missive shift from conventional military doctrine.
Some countries are probably closer to it than others. The US has great tech and resources to do something like that. China makes all the drones and has experience with AI targeting of their citizens with the social credit stuff.
Russia is retarded.
I don't think anyone is ready for "swarms" yet. I believe you will see a slow evolution of "drones that directly protect tanks" or "drones that assist infantry do stuff like clear rooms and tunnels". To once the tech becomes proliferated then you might see the idea come about of swarms.
Obviously you need to keep in mind things that counter this strategy. Such as active camo systems, jammers, and hacking.
The idea a single operator could send a swarm is pretty unlikely. I imagine the tech would be resource heavy. Also, the value of sending a human in person to do things will probably always be there.
You need to also give consideration to the logistical aspect of this. Where do the batteries come from, do they suicide or re-arm? Does weather effect performance? Also the enemy can always see where the drones originate from if they have the tech.
The idea of swarms sounds cool, but the manufacturing capacity, the tactics and strategy, alongside the usefulness/uselessness ratio of this is not there yet.
Seeing these videos of how effective the ukranian nade dropping drones are for how little they cost, the idea of a computer controlling some cheap ardupilot drone swarm, all equipped with a couple of nades, just seems like the molotov coctail of 21st century
Immagine those ukrainian drone videos, but instead of 1 or 2, suddenly 100 drones show up and start lobbing nades.
>6 years ago we showed this publicly
Those are only surveillance, I think op means offensive drones.
>hundreds of drones loudly swarming around a tiny area
>surveillance
Only thing a Brimstone doesn't do is come home if it doesn't find a target, so we're already there and it's more cost effective to just have the drones not bother with being recoverable.
Pretty close, but you face some big problems for it.
1: target identification. If you manage to get an ai to consistently recognize humans, how will you teach it to differentiate between friendly, enemy, and civilian? You'll need to or You'll get major issues. Current AI image recognition is pretty good, but not perfect.
2: data structure. Are you going to have the drones be slave to a proper server bank doing the calculations or are they to be autonomous? The latter option is both more dangerous and way, way more costly, as you need to pack them with enough brain and battery to do so efficiently over time. The former is susceptible to interference.
3: air time.
Current small rotary drones used for combat do not have a great amount of time in the air before running out of charge. You'd need to improve this without ruining what performance statistics they have.
Isnt one of the biggest bonuses of drones that they are small and hard to detect and effectivly combat? Why put a whole lot of them in a swarm and make it wasier to detect and destroy them?
Well, the US has signals jamming so .....it's usless
What if you're not fighting the US?
Laser can't be jammed and their are other techniques also. The issue I see is that swarms imply lots of small drones. You need larger drones for the range and explosive power. This is more like the idea of that short film killbots. They were small because they had just enough explosives to kill a person with an impact to their head.
>figure out miniature nukes
>strap one to a drone
>sink an aircraft carrier
https://youtube.com/shorts/Z_Ib3wRFGPs
Using them as swarms? That's hard. A kill mission with an AI detection algorithm at the last mile? That's easy, and super easy if you have a thermal. I'd be surprised if this isn't being used in Ukraine RN.
5-10 years.
You’ll see first demonstrators in 2-3 years.
t. works on such a project
It’s a fixed wing though.
Also in about the same time frame you can expect active optical camouflage though its use will be limited because of how it works and what compromises it makes. Basically only small, low flying, fixed wing UAVs with specific airframe design.